Salzer and Hughes on Bristlecones

A reader writes in:

I think Steve and the other CA readers will be interested in the new article by Matthew Salzer and Malcolm Hughes (of MBH fame) entitled “Bristlecone pine tree rings and volcanic eruptions over the last 5000 yr” in the latest issue of Quaternary Research, “> available for free.

I don’t have time to review this right now but look forward to it.

Here’s the reader’s summary:
The article begins quite straightforwardly with the observation that “Over the last millennium the agreement between the tree-ring data and volcano/ice-core data is high: years of ring-width minima can be matched with known volcanic eruptions or ice-core volcanic signals in 86% of cases’. It actually strikes me as quite plausible that cold snaps could immediately and rather unambiguously retard tree-ring growth, unlike warming which usually would happen slowly over many years and with many complicating factors.

The introduction is also clear: ” Two primary factors are thought to have forced much of late Holocene variation in climate prior to industrialization: solar output and volcanic eruptions (Free and Robock, 1999, Crowley, 2000 and Shindell et al., 2001). While there is some debate regarding which of these forcings has played the dominant role (Shindell et al., 2003), there is little doubt that volcanism affects climate. Large explosive eruptions inject great quantities of sulfur compounds into the stratosphere, which combine with water to produce sulfuric acid aerosol (Rampino and Self, 1982). This injection changes the radiative balance by increasing absorption and reflection of incoming short wave radiation by stratospheric aerosols, and generally has a cooling effect on climate (Lacis et al., 1992, Minnis et al., 1993 and McMormick et al., 1995)”.

However, I get nervous when I read

“Standardization is a basic procedure in dendrochronology that is designed to remove long-term non-climatic factors associated with increasing tree age and tree size from individual time series of ring-width measurements (Fritts, 1976 and Cook et al., 1990). The standardization procedure involved the fitting of a line or curve to the individual sample ring-width series using the program ARSTANL (Cook et al., 1985), (version 6.04P; also available at: and dividing the raw data by the fitted curve. Due to the open non-competitive nature of high elevation bristlecone pine stands, a modified negative exponential curve, a straight line of negative slope, or a horizontal line were used in the standardization. To create a mean site chronology, the annual standardized indices of tree growth were averaged. This process was repeated for each of the five upper forest border chronologies used in the study. Mean segment length was 492 yr, so that this standardization procedure should retain variability on time scales up to 164 yr (Cook et al., 1995), but as time scales lengthen beyond 164 yr the proportion of variability retained declines.

The variances of the chronologies were adjusted to remove variance bias as a result of sample size (Osborn et al., 1997; pp. 90–92, equations 4–6) and the chronologies normalized using the mean and standard deviation. The resulting normalized index chronologies consist of a variance-adjusted average of many individual samples from living trees and from dead remnant material. The average correlation between the five chronologies over their common intervals is 0.41 (n > 2230 yr, all significant at p < 0.0001) despite an average distance between sites of over 440 km. The five chronologies were then averaged to form a single time series. This series was also adjusted to remove variance bias (Osborn et al., 1997) and normalized. The resulting regional tree growth time-series is a high elevation chronology of upper forest border tree growth from 3000 BC to AD 2002 with a variance that does not depend on the size of the sample. The chronology minimizes any single site idiosyncrasies and maximizes the signal common to all five sites (Fig. 2), (hereafter referred to as the HI5 chronology). It should be noted that due to the nature of chronology building, particularly the use of overlapping series with a mean length of approximately 500 yr, millennial-scale variability would not be retained in the HI5 chronology".

And the picture on page 60 looks awfully familiar, even if it’s a tree-ring index rather than temperature. Anyway, I hope Steve and some of the CA readers will have a look. There are a few other articles in this free issue that are also interesting.

Ref: Bristlecone pine tree rings and volcanic eruptions over the last 5000 yr, Matthew W. Salzer and Malcolm K. Hughes, Quaternary Research Volume 67, Issue 1 , January 2007, Pages 57-68, doi:10.1016/j.yqres.2006.07.004


  1. Posted Jan 21, 2007 at 4:55 PM | Permalink

    Here’s the Figure 2 mentioned above.

  2. Steve McIntyre
    Posted Jan 21, 2007 at 10:12 PM | Permalink

    This article doesn’t discuss the 20th century growth pulse at all. It does not include updated Sheep Mountain information although it uses Sheep Mountain. Sheep Mountain was updated in 2002; a notice that an updated chronology was reported in April 2004 here . Yet the 2007 publication continues to use obsolete data. Can you imagine a business updating its financial records and then reporting only its results up to 2001.

    I made the prediction long ago that the Sheep Mountain update will not show elevated growth through the warm 1990s and early 2000s. The present article does not settle the prediction – other than the matter is avoided by Hughes one more time.

  3. Posted Jan 22, 2007 at 9:57 AM | Permalink

    The accuracy of the ice-core dates prior to ~0 BC is highly disputable. A good reference for this is J. Southon [Radiocarbon 46: 1239–1259 (2004)].

    I’ve done some analysis that supports and extends Southon’s work. I hope to publish it, but getting all the data is problematic.

    In my view, the ice-core guys are heroes of science for gathering their data—spending years out on the Greenland ice cap. And their data is extremely valuable.

    But I strongly dispute their ability to analyze data. Salzer & Hughes cite the work of ice core researchers Hammer et al.&nbsp[2003] on some volcanic ash; they also cite a paper of mine that criticizes Hammer et al. In fact, the main thing that my paper does is point out that Hammer et al. do not understand the concept of standard deviation. Worse, I had discussions about that with the authors prior to the publication of their paper, and I only published mine when they refused to reconsider.

  4. Posted Jan 22, 2007 at 11:56 AM | Permalink

    Despite the obfuscation the issue here seems tivial. If I have a time series that varies randomly and I divide the data values by a constant at every point (ie. a straight line) I will get a unbiased transform. That is to say if this is part of a smoothing procedure I will not have introduced anything into the analysis.

    However here the author states:

    a modified negative exponential curve, a straight line of negative slope, or a horizontal line were used in the standardization.

    Such a procedure must give the resulting smoothed series a rising character. This seems to be the exact same kind of procedure used in MBH98 with PCA. There, as I remember, the authors used a different mean value for recent measurements resulting in a “hockey stick” shape for the smoothed graph. Here the author creates a slope in the results by deviding the original measurements by a slope.

  5. Steve McIntyre
    Posted Jan 22, 2007 at 1:59 PM | Permalink

    #4. This adjustment is done at the level of individual trees to remove early growth trend. Resulting chronologies do not necessarily have a HS-shape. I agree that there is risk in this standardization process no matter how hard they try and the bias is to HS-ness. But I don’t think that that’s the issue with bristlecones – their ring widths increase in the 20th century no matter how you do it. Whether the increase is due to temperature is a different question.

  6. henry
    Posted Oct 31, 2007 at 12:39 PM | Permalink

    Eli Rabbet is just now getting around to this, so in his honor, I RTFR.

    “While ring width in bristlecone pine at the upper limit of its distribution may be strongly influenced by temperature or a temperature-related condition such as growing season length, in this semi-arid region drought may also influence growth, even at high elevation. The HI5 chronology was subjected to two separate Superposed Epoch Analyses (SEA) in an attempt to separate these two influences…

    The SEA analysis period was AD 1595–1898; 20th century data were eliminated so as to avoid potential biases associated with industrial-era atmospheric contamination (LaMarche et al., 1984, Graybill and Idso, 1993 and Hughes and Funkhouser, 2003).”


    “Time-series plot of regional high elevation tree-ring index from 3000 BC to AD 2002 (HI5 chronology-variance-adjusted normalized mean of five subalpine bristlecone pine chronologies).

    Bristlecones again. Once again, does this tie back to the original MBH study?

    “Mann et al., 1998 M.E. Mann, R.S. Bradley and M.K. Hughes, Global-scale temperature patterns and climate forcing over the past six centuries, Nature 392 (1998), pp. 779–787.”

    Guess so.

  7. bender
    Posted Nov 3, 2007 at 10:31 AM | Permalink

    From the above:

    It has long been suspected that upper-treeline ring widths are recording variability in temperature or a temperature-related variable such as growing season length (LaMarche, 1974; LaMarche and Stockton, 1974). This has been difficult to demonstrate with modern meteorological data, however, possibly due to a reduction of ring-width sensitivity to temperature.

    [Emphases mine.]

    Decoding the italicized phrase … this is more evidence that these guys understand that tree ring reponses are nonlinear and synergistic, and therefore unreconstructable through linear, univariate correlative/regressive models.

    … An inference that might not obvious from a superficial reading.

    NB: They should have provided a citation for the second statement. McIntyre’s work at CA being the most obvious choice.

%d bloggers like this: